# How a $40M Manufacturer Captured 25 Years of Expertise Canonical: https://granular.to/blog/manufacturer-captured-25-years-expertise Published: 2026-05-23 Updated: 2026-05-23 Author: Trey Category: Case study Tags: manufacturing, knowledge-management, ai-agents, case-study > When a $40M precision-parts manufacturer's senior estimator announced retirement, five binders and a year of shadowing had already failed. Four weeks of structured AI knowledge capture preserved 25 years of quoting judgment and cut turnaround from three days to same-day. > **TL;DR:** A $40M precision-parts manufacturer was six months from losing their senior estimator and the 25 years of quoting judgment he carried in his head. Traditional documentation, including binders, shadowing sessions, and SOPs, had failed. A 4-week AI engagement captured his decision logic into a queryable tool: structured extraction sessions in week one, build and iterate in weeks two and three, calibration alongside the expert in week four. New estimators now quote within 8% of the veteran's accuracy. Quoting turnaround dropped from three days to same-day on standard jobs. The problem showed up in a Monday morning staff meeting. The owner of a 60-person precision-parts shop announced that their senior estimator, call him Dave, had given notice. Six months out. Thirty-one years on the shop floor, twenty-five of them quoting complex machined components for aerospace, defense, and industrial OEMs. The VP of Operations had tried to solve this before. Three years earlier, she had Dave document his process in binders. Two years before that, she had junior estimators shadow him four days a week. Neither stuck. The binders were accurate the day they were printed and wrong six months later when tolerances tightened or material costs shifted. The shadowing produced estimators who could copy Dave's answers but could not replicate his reasoning when a new part type showed up. Dave's quoting accuracy was not about memory. It was judgment, specifically, the ability to look at a customer drawing and weigh cycle times, scrap risk, tooling amortization, setup complexity, and historical win rates for that customer in one mental pass. That is not a procedure. It is a decision model built from thousands of jobs. With 1.9 million manufacturing jobs projected to go unfilled through 2033, according to a joint Deloitte and Manufacturing Institute study, and 65% of manufacturers citing talent attraction and retention as their primary challenge, Dave's situation is not unusual. It is the norm. What made this shop different was deciding to do something about it before the door closed. ## Why Documentation Always Fails This Problem Binders fail because they capture what, not why. A new estimator reading Dave's binder learns that titanium aerospace brackets get a 22% scrap buffer. They do not learn that the buffer jumps to 35% on parts with blind holes deeper than 4xD, or that it drops back to 18% when the customer supplies pre-certified bar stock. That conditional logic, built from watching five bad runs over twelve years, never makes it into an SOP. Shadowing fails because it is asymmetric. Dave knows which decisions matter. A junior estimator watching over his shoulder does not know which details to capture and which to ignore. They write down everything and learn nothing transferable. The pattern holds across manufacturing. Research from the Association for Talent Development found that only 7% of industrial companies consistently capture knowledge from departing experts, even though 83% of C-suite leaders rate knowledge loss as mission-critical or a strong concern. The gap is not awareness. It is method. ![Annotated shop drawing with handwritten tolerance notes on a steel surface, shallow depth of field](/images/blog/manufacturer-captured-25-years-expertise-annotated-drawing.jpg) ## What a 4-Week Capture Engagement Actually Looks Like The shop engaged Granular for a fixed-scope knowledge capture project. Here is how the four weeks ran. **Week 1: Extraction sessions.** The work began with structured interviews, not open-ended conversations. The goal was to surface Dave's decision trees, not his general opinions. Facilitators asked specific questions: "Walk me through the last job where you came in above your initial estimate. What did you miss?" "What's the first thing you look at on a drawing that tells you this job will be hard to price?" Three two-hour sessions over five days. Every response was recorded, transcribed, and tagged by decision type: material selection, setup time, scrap risk, customer history, tolerance risk. **Weeks 2 and 3: Build and iterate.** The tagged transcripts fed a knowledge graph. Granular structured Dave's conditional logic into queryable rules: "If tolerance band is under 0.001 and feature is internal, then setup complexity flag triggers." Junior estimators used a prototype version daily on live RFQs, documenting every instance where the tool's suggestion differed from Dave's verbal answer. Those divergences drove refinement. By day 18, the tool's suggestions matched Dave's independent quotes on 74% of parts without further input. **Week 4: Calibration.** Dave and the two junior estimators worked the same RFQ queue simultaneously, comparing outputs in real time. Disagreements became teaching moments. Not "Dave says X," but "here is the reasoning path the tool used and here is where Dave's judgment diverges." By the end of week four, alignment on standard jobs hit 92%. ## The Numbers Six Months Later The owner tracked three metrics from day one of Dave's departure. **Quoting accuracy.** New estimators are quoting within 8% of Dave's historical win rate on like-for-like parts. Before the engagement, junior estimators were 23 to 31% off on complex assemblies. **Turnaround time.** Standard jobs, defined as parts the shop has run before in similar materials and tolerances, now quote same-day. Previously, Dave was the only person who could approve those quotes, creating a three-day average wait when he was occupied on larger bids. **Win rate on re-quotes.** When customers came back for updated pricing on existing parts, the shop was previously inconsistent. Prices shifted because whoever picked up the quote applied different logic. That inconsistency has dropped. The tool maintains the prior reasoning for every part and flags when a re-quote deviates from historical patterns by more than 10%. These results map to a pattern documented in manufacturing research. A Workplace Intelligence analysis of knowledge transfer programs found that systematic capture reduces new-hire ramp time by 50% and improves first-time accuracy rates by 15 to 25% on complex technical tasks. The numbers at this shop landed toward the top of that range because the engagement started while Dave was still available to calibrate. ## What the Owner Would Have Done Differently When asked what he would change, the owner said he would have started 18 months earlier, not six. At six months, the team had enough time to capture Dave's logic for the core job families that drove 80% of revenue. But there were edge cases, specifically, quoting military-spec jobs with unusual surface treatment requirements, where Dave's intuition ran deep and the capture window was too short to fully surface it. That reflects a finding from Automation.com research on manufacturing knowledge transfer: knowledge capture started fewer than 12 months before retirement recovers roughly 60 to 70% of actionable expertise. Starting 18 to 24 months out consistently recovers more than 85%. The difference is not more interviews. It is more time for the tool to surface gaps through live use and for the expert to fill them before the deadline pressure arrives. The shop is now treating the remaining 20% as a living document. Dave agreed to a six-month consulting arrangement, two hours per month, specifically to answer questions the tool flags as outside its confidence threshold. That structure keeps the knowledge pipeline open without requiring full-time re-engagement. ![A steel machine dial gauge sitting on a precision surface plate under fluorescent shop lighting, no faces visible](/images/blog/manufacturer-captured-25-years-expertise-precision-gauge.jpg) ## What Makes This Different From an AI Chatbot It is worth being precise about what this tool is and is not. It is not a general-purpose AI chatbot pointed at a pile of documents. It is a structured knowledge graph built from Dave's specific decision logic, connected to the shop's historical job data, and queryable in the context of an active RFQ. The distinction matters because the failure mode of general-purpose AI tools in manufacturing quoting is confident wrongness. A model trained on generic manufacturing content will answer questions about titanium machining with broad accuracy and specific inaccuracy, because it does not know this shop's machines, this shop's scrap history, or this customer's tolerance tendencies. Dave's tool answers from Dave's data, with traceable reasoning that the estimator can audit. This is also why the build-vs-buy question matters for shops in this situation. Off-the-shelf quoting software will systematize the workflow but it will not capture the judgment layer. Understanding that distinction before starting is the difference between a tool your team uses and a tool that sits on a server. For a full breakdown of where custom AI builds make sense versus packaged software, the [build-vs-buy guide for mid-market operators](/blog/build-vs-buy-ai-mid-market-guide) covers the decision framework in detail. ## The Real Question Is Timing The headline outcome, 8% accuracy, same-day turnaround, is meaningful. But the more important number is six months. That was how close this shop came to Dave walking out the door without a transfer in place. A Deloitte and Manufacturing Institute report found that nearly one-third of the U.S. manufacturing workforce was over 55 years of age as of 2022. That cohort carries the institutional knowledge base of most mid-size shops. Waiting until a retirement notice arrives is not a strategy. It is a bet that the gap will be small enough to survive. If you have a Dave on your team, the [full playbook on capturing tribal knowledge before key people leave](/blog/capture-tribal-knowledge-before-key-people-leave) walks through the framework for identifying who holds critical decision logic, how to prioritize capture by business impact, and what a realistic engagement timeline looks like. The shop in this case study did not win because they had a sophisticated AI strategy. They won because they started two weeks after Dave's notice instead of two weeks before his last day. That gap bought enough time to do the work right. --- If your shop is within 18 months of a key retirement, or if you have already lost someone and are feeling the gap, [book a discovery call with Granular](/). We scope knowledge capture engagements in a single call and can tell you within 30 minutes whether a structured program makes sense for your situation. --- ## Keep Reading - **[Capture Tribal Knowledge Before Key People Leave](/blog/capture-tribal-knowledge-before-key-people-leave)** — The step-by-step playbook for identifying who holds critical knowledge, how to prioritize by business impact, and what a realistic capture timeline looks like. - **[Build vs. Buy AI: The Mid-Market Guide](/blog/build-vs-buy-ai-mid-market-guide)** — How to decide whether off-the-shelf software or a custom AI build is the right call for your specific operational problem.